Avoid double-work when publishing blocks

[peterschwarz]

This is a significant and relatively easy performance enhancement.

Currently, any node publishing a block does the following (roughly):

a) process transactions that will be in the block
b) create the block, publish it
c) process transactions to verify the block
d) update chain head

For step (c), we can use a cached result of (a). This will cut down the validator CPU usage significantly and should nearly double the transactions-per-second of the validator.